Thresholding methods review for microcalcifications segmentation on mammography images in obvious, subtle, and cluster categories

Abstract

Microcalcifications are the earliest sign of breast carcinoma. Their typical size is about 1 mm, which is why it is difficult to detect for an expert. Therefore, a tool that eases their visualization becomes relevant. Segmentation gives the candidate areas that could contain microcalcifications. A preprocessing step can improve segmentation performance but the algorithm becomes database dependent. This paper compares four commonly used thresholding techniques to segment mammography images having sections divided in three groups: obvious, subtle and clusters; due to their microcalcification contents. The purpose of this paper is to show what technique has a better performance in special relation with mammography images. Best performers are Entropy (68.8%), and Intermodes (50.9%), but further research is needed to improve performance on subtle and cluster microcalcifications considering non-bimodal histograms.